- In this study we demonstrate how we interconnect three different ontologies, the BioAssay Ontology (BAO), LINCS Information FramEwork ontology (LIFEo), and the Drug Target Ontology (DTO). The three ontologies are built and maintained for three different projects: BAO for the BioAssay Ontology Project, LIFEo for the Library of Integrated Network-Based Cellular Signatures (LINCS) project, and DTO for the Illuminating the Druggable Genome (IDG) project. DTO is a new ontology that aims to formally describe drug target knowledge relevant to drug discovery. LIFEo is an application ontology to describe information in the LIFE software system. BAO is a highly accessed NCBO ontology; it has been extended formally to describe several LINCS assays. The three ontologies use the same principle architecture that allows for re-use and easy integration of ontology modules and instance data. Using the formal definitions in DTO, LIFEo, and BAO and data from various resources one can quickly identify disease-relevant and tissuespecific genes, proteins, and prospective small molecules. We show a simple use case example demonstrating knowledge-based linking of life science data with the potential to empower drug discovery.
I. INTRODUCTION Big data are ubiquitous in business, technology and science. Life science research data are no exception. However, the nature of research data, in particular in the life sciences brings additional challenges due to broad diversity of data types and formats, the quick evolution of knowledge and advancements in technologies to generate data. Despite large investments in information systems in the pharmaceutical industry and nonprofit research organizations, the difficult problem of describing, organizing, integrating, analyzing diverse, fast evolving and large scale data in the context of biological knowledge remains a critical and not fully solved challenge. In this study we demonstrate in a simple case study how to represent and organize such data better by using Semantic Web technologies. Although this approach is not novel, we contribute by leveraging three ontologies developed in our group and that are largely aimed at addressing different aspects of drug discovery data.
The BioAssay Ontology (BAO) [3] has been developed to
formally describe knowledge of chemical biology assays and
screening results using Description Logic (DL) [14] and OWL
(OWL2.0) [17]. The first version of BAO [9] focused on High
Throughput Screening (HTS) assays and contained
descriptions of many assays from PubChem. BAO since
evolved to better integrate with other ontologies and better
align with established upper level models and improve
usability. BAO was also extended to support profiling assays
such as those in LINCS [
The concepts for BAO ontology are either created by our
group, or extracted from external ontologies and used with
their own URIs. LIFEo formally describes data generated in
the LINCS project’s Data and Signature Generating Centers
(DSGCs). Finally, for DTO we formally describe drug target
data that are the focus of the IDG Project. We further use
public databases, such as UniProt [
We used Protégé [
III. RESULTS
BAO [3] was designed and implemented to axiomize
knowledge about bioassays. As the content expanded with the
addition of LINCS assays, an architectural change was
implemented to the ontology so that it can maintain its core
while importing external ontologies for existing information.
Current version of BAO has >3300 classes, >420,000 axioms
and 165 object properties.
Briefly, the implemented modular architecture divides the
ontology into layers, starting with the vocabularies, followed
by modules with BAO-native axioms, and finally, different
views of the ontology can be created by combinations of
modules that can contain the native as well as the external
axioms. An important feature of this modularization is that it
allows to create a BFO-founded version for ontology authoring
and integration with other resources, but also a BAO-native
version for users; since most users are not familiar with BFO
terms. In addition to the ontology architecture of BAO, we
aimed to standardize the assay descriptions by creating
metadata and design patterns for the formal definitions. LINCS
assays were axiomized in BAO using the model previously
described [
The Library of Integrated Network-Based Cellular
Signatures (LINCS) project aims to create a network-based
understanding of biology by cataloging changes in gene
expression and other cellular processes that occur when cells
are exposed to a variety of perturbing agents. LINCS aims to
use computational tools to integrate this diverse information
into a comprehensive view of normal and disease states that
can be applied for the development of new biomarkers and
therapeutics [
To accomplish this, we have formalized LINCS assays in BAO. The systems biology nature of LINCS data required a new model we called the LIFE ontology. LIFEo is an application ontology designed to handle the different biological molecules and model systems (in particular cell lines and cells), their relationships to other concepts, such as disease and tissue and assays and their roles. LIFEo contains >49000 classes, >132,000 axioms and 62 object properties (including direct and indirect imports from BAO, DTO and other, external ontologies). By using a modular approach, LIFEo is aiming to create a useful model of how the different metadata components in LINCS align across the entire project.
The first version of LIFEo supported eight assays, namely:
KINOMEscan, KiNativ, Cue Signal Response, P100,
2-ColorApoptosis, 3-Color-Apoptosis, Cell Cycle State, and Cell
Growth Assays [
Although LINCS assays are diverse with respect to assay technology, the detection method, model systems, and metadata entities, the main point of LIFEo is to provide an integrative model that facilitates context-specific analysis by formalizing the most important relationships.
In addition to the gene and protein modules of the LIFE ontology, we have a module for cells that are used as model systems in the various assays. They are grouped into four main categories: stem cells, primary cells, differentiated cells, and cell lines. Cell lines then are grouped by using the organs from which they were derived using UBERON.
DTO is being created as part of the IDG project. An important
goal of the IDG project is to catalyze the development of
chemical probes and drug development entry points for
understudied, yet relevant protein targets in the four most
commonly targeted protein families (G-protein coupled
receptors (GPCR), nuclear receptors, ion channels, and
kinases) by integrating all available information and making it
available as actionable knowledge. The current version of
DTO consists of asserted class hierarchies of the ~1800
protein targets, > 13,000 classes and > 214,000 axioms.
DTO is designed to work with other ontologies, such as BAO
and thus can be used to describe proteins in LINCS assays.
DTO content is being curated from various sources and the
details of the development of DTO will be described
elsewhere. DTO content is further annotated and linked by
various ontologies. To facilitate the construction of DTO, we
wrote various scripts using Java to retrieve information from
databases and ontologies. These databases include UniProt
and NCBI databases for ENTREZ IDs for the genes, and
ChEBI [
We retrieved the proteins, with their tissue and disease relationships with the confidence scores that are given to the relationships. We put this data into our database and use this information while creating the ontology's axioms that refer to the probabilistic values of the relationships.
Drug Target Ontology (DTO) uses various external databases and ontologies to retrieve information. The information is retrieved via web-based applications and inhouse-built scripts. The data that is used to build DTO is then housed in an internal database. To facilitate ontology development and maintenance, such as frequent updates and synchronization to other data sources, we use Java, OWL API and Jena to build modules of the ontology in a semi-automated. The details of the specific modularization architecture are shown in Figure 3.
In contrast to BAO, which is primarily constructed manually by experts formalizing axioms, DTO integrates lots of information from different resources. We therefore separated a further module category built using only automated scripts. These are imported into modules that incorporate expert-built axioms. This way, updates from the database will not overwrite expert-modeled content.
First, we determine the abstract horizon between TBox and ABox. Tbox contains modules, which define the conceptualization without dependencies. These modules are self-contained and well-defined with respect to the domain and they contain concepts, relations, and individuals. We can have n of these modules.
Second, once the n modules are defined, the modules with axioms that can be generated automatically are created. Those modules have interdependent axioms. At this level one could create any number of gluing modules, which import other modules without dependencies or with dependencies. It also is self-contained. This means that there is no outside term or relationship in the files.
Third, this level contains axioms created manually; however the axioms generated are independent and selfcontained. The manual modules are an optional level and they inherit the axioms created automatically. A good example of axioms that may be seen in this level are axioms for protein modifications and mutations, which have been challenging modeling questions. At this level, the self-contained DTO_core is also generated with the existing modules.
Fourth, at this level we can design modules that import
modules from our domain of discourse, and also from third
party ontologies. Third party ontologies could be large,
therefore a suitable module extraction method (e.g. Java
programs using OWL API and Jena) can be used to extract
only part of those ontologies (vide supra). We would model
this in the DTO_complete level. We can have one
DTO_complete file or multiple files, each may be modeled for
a different purpose, e.g., tailored for various research groups.
Once these ontologies are imported, the alignment takes place.
The alignments are defined for concepts and relations using
equivalence or subsumption DL constructs. The alignment
depends on the domain experts and/or cross-references made in
the ontologies. For DTO, the most significant alignment made
is between UBERON [
Fifth, release the TBox based on the modules created from the third phase. Depending on the end-users, the modules are combined without loss of generality. With this methodology we make sure that we only send out physical files that contain our (and the absolute necessary) knowledge.
Sixth, at this level, the necessary modules ABoxes are created. ABoxes can be loaded to a triple store or to a distributed file system in a way that one could achieve pseudoparallel reasoning.
Seventh, at this level we define views on the knowledge base. These are files that contain imports (both direct and indirect) from various TBoxes and ABoxes modules for the end-user. It can be seen as a view, using database terminology.
Since BAO, LIFEo, and DTO have been constructed using a modular approach, we are able to create different views that would help to integrate and query relevant data, for example in the drug-discovery domain. We extracted the LINCS assays from BAO by using Jena [6], and OWL API, used the cell line module from LIFEo, and the targets from Drug Target Ontology (DTO) in order to query the following use case.
What are the kinases used in the LINCS assays that measure protein binding and have strong evidence for being associated with cancer? Further, what are the relevant compounds targeting these kinases and are therefore relevant for the disease? What other data support the compound-disease association? The generic example cancer, is meant only for illustrative purposes. This query aims to retrieve assay specific proteins based on the assays of interest. It works in two parts. In the first part we used the molecular function that is measured (i.e. protein binding) to infer the bioassays of interest. This information is formally described in BAO. We then identified the kinases used in these assays using the LINCS data axioms related with kinases in LIFEo. Finally by using DTO, we get the intersection of this subset of kinases with the kinases that have strong evidence for associations with cancer. The disease and tissue information related with different genes and proteins is formalized in DTO as described above. We further analyzed the results by using the compound data in the LIFEo. We queried the compounds used both in KINOMEscan (KS) and KiNativ (KN) assays.
Assay
s KS& KN KS& KN KS KS& KN KS& KN KS Figure 4 Illustration of the example query using the three ontologies We combined the resulting kinases of Query1 with the 22 compounds. Table 2 shows the specific kinases that were targets of the same assays as the 22 compounds used both in KINOMEscan and KiNativ assays.
In summary, assays with their molecular functions of interest are axiomized in BAO. Kinases have assay related axioms in LIFEo, which we retrieve as the second step in the query. We then explore more about the proteins by using the axioms related with their associated disease information from DOID [8] encoded in the DTO. As cell lines are linked to diseases, compounds can further be identified based on the growth inhibition assays.
Our results showed us that with the three ontologies, BAO, LIFEo, and DTO, we were able to connect different data types and content related to drug-discovery data. The uniform architecture along with the complex and sequential modeling templates we use for the diverse types of data, allows us to combine different modules and create different views in order to reach the components of interest faster.
IV. DISCUSSION Here we presented three ontologies built for three related, yet different projects, and how they can work together in queries crossing several concepts important for drug discovery. This is facilitated by the similar modular architectures of the ontologies, which enable their integration of diverse information into a triple store.
BAO has been developed to formalize complex chemical
biology assays, such as HTS assays, which are one of the
primary methods to identify novel entry points for drug
discovery projects. BAO facilitates re-use of this data. LIFEo
provide a simple model to address the systems biology
aspects, specifically relations of disease model systems,
tissues, protein targets, small molecules and assays. DTO
describes drug targets formally and integrates information
from many sources. All ontologies utilize external ontologies,
which serve as an integration point, such as disease and tissue.
BAO was used in the BioAssay Research Database (BARD)
software system [19] and it is used in several projects and
organizations [
DTO provides a formal classification of four protein families based on function and phylogenetic and describes their clinical classifications and relations to diseases and tissue expression. DTO is already used in the IDG main Portal Pharos (https://pharos.nih.gov/) and the TinX software application (http://newdrugtargets.org/) to prioritize drugs by novelty and importance. DTO is publicly available at http://drugtargetontology.org/, where it can be visualized and searched.
We have illustrated how DTO, LIFEo, and BAO and included external ontologies are used to describe, integrate, and query drug discovery related data. We are also in the process of integrating these knowledge models with the recently released LINCS Data Portal (http://lincsportal.ccs.miami.edu/). For the purpose of this paper, we have integrated only a part of the available LINCS data in a local triple store to demonstrate the basic concept of our approach of integration. Much more work is required to fully integrate and model all LINCS data. As we expand the LINCS and DTO knowledge models, we can construct more complex queries. A particular goal is to enable the context-sensitive integration and querying of data. We will also integrate further ontologies for example the Cell Line Ontology (CLO) to formalize LINCS cell lines.
We continue to develop BAO and DTO to maximize their utility for the research community. We are constructing a more advanced LINCS MetaData Ontology towards the goal of a comprehensive systems-based model of LINCS signature and drug discovery data.
ACKNOWLEDGMENT This work was supported by grants U54CA189205 (Illuminating the Druggable Genome Knowledge Management Center, IDG-KMC) and U54HL127624 (BD2K LINCS Data Coordination and Integration Center, DCIC). The IDG-KMC is a component of the Illuminating the Druggable Genome (IDG) project (https://commonfund.nih.gov/idg) awarded by the NCI. The BD2K LINC DCIC is awarded by the National Heart, Lung, and Blood Institute through funds provided by the trans-NIH Library of Integrated Network-based Cellular Signatures (LINCS) Program (http://www.lincsproject.org/) and the trans-NIH Big Data to Knowledge (BD2K) initiative (http://www.bd2k.nih.gov). Both IDG and LINCS are NIH Common Fund projects. membranes.
Vol. 507.
Kibbe, Warren A., et al. "Disease Ontology 2015 update: an expanded and updated database of human diseases for linking biomedical knowledge through disease data." Nucleic acids research 43.D1 (2015): D1071-D1078.
Schürer, Stephan C., et al. "BioAssay ontology annotations facilitate cross-analysis of diverse high-throughput screening data sets." Journal of biomolecular screening 16.4 (2011): 415-426. [10] Gene Ontology Consortium. "Gene ontology consortium: going forward."Nucleic acids research 43.D1 (2015): D1049-D1056. [11] Brinkman, Ryan R., et al. "Modeling biomedical experimental processes with OBI." Journal of biomedical semantics 1.1 (2010): 1. [12] Tsarkov, Dmitry, and Ian Horrocks. "FaCT++ description logic reasoner: System description." Automated reasoning. Springer Berlin Heidelberg, 2006. 292-297. [13] Dumontier, Michel, and Melanie Courtot. "Proceedings of the 8th
International Workshop on OWL: Experiences and Directions." (2011). [14] Baader, Franz. The description logic handbook: Theory, implementation and applications. Cambridge university press, 2003. [15] Hitzler, Pascal, Markus Krotzsch, and Sebastian Rudolph. Foundations of semantic web technologies. CRC Press, 2009. [16] Gruber, Thomas R., Nicola Guarino, and Roberto Poli. "Formal ontology in conceptual analysis and knowledge representation." Chapter" Towards principles for the design of ontologies used for knowledge sharing" in Conceptual Analysis and Knowledge Representation (1993). [17] W3C Owl Working Group. "{OWL} 2 Web Ontology Language
Document Overview." (2009). [18] Xiang, Zuoshuang, et al. "OntoFox: web-based support for ontology reuse."BMC research notes 3.1 (2010): 175. [19] Howe, E. A., et al. "BioAssay Research Database (BARD): chemical biology and probe-development enabled by structured metadata and result types."Nucleic acids research (2014): gku1244. Fund,
Based on all the reviewers’ request, we added pages with larger figures. We couldn’t see a way to add supplementary materials. Figure 2 Figure 3